So chatten Sie mit Ihren Dokumenten

Most questions about documents span more than one file. What did the original brief say about scope, and did the second round of research change it? Which of the three policy drafts mentions remote work? A single document can be read end to end, but a folder of twenty takes an afternoon, and the answer often depends on how two of them relate.

This guide covers the usual ways people put questions to their documents, where each one starts to strain, and how to hold a back-and-forth conversation with a set of documents in Fabric, with every answer pointing to its source.

How people usually ask questions of their documents

The oldest method is opening each file and pressing Find. It works when you know the exact word the author used. It works less well for questions like “what were the main objections”, where the answer is spread across paragraphs that never use the word objection.

Many people now upload files into ChatGPT or Claude and ask there. For a handful of files this is quick and the answers are good. The friction shows up over time: each new conversation needs the files again, updated versions have to be re-uploaded, and the chat lives in a separate place from where the documents are kept. There’s more on that pattern in the guide to using ChatGPT with your own files.

Others use the assistant built into their word processor. That works well on the document in front of you. When a question needs a PDF, two Word files and a set of meeting notes at once, people tend to fall back on copying passages between windows and keeping track of which quote came from where.

Step 1: Get the documents into Fabric

Drag the files into the Fabric web or desktop app, paste them, or click Upload a file. Uploading handles large numbers of files at once, so a project folder can come across in one go. Fabric reads PDFs, DOCX, TXT, Markdown, HTML and JSON, and you can paste text straight in to create a note.

If your documents live in Google Docs or Notion, a connection brings them in. Google Drive syncs both ways, so edits made in either place carry across. Connections need a paid plan. If you work mostly from Finder or File Explorer, the desktop app mounts Fabric as a drive and documents can be moved in like any other folder.

Step 2: Group the documents you want to talk to

The assistant can search your whole workspace, but a conversation is sharper when it has a defined set of files. Put the documents for one project, client or topic in a folder. Fabric’s folders, spaces and tags are all optional, so this can be as light as one folder per piece of work.

Include the material that gives the documents context: the email thread that explains why a draft changed, the notes from the meeting where it was discussed, the earlier version. The assistant analyzes everything in the folder, so anything you add there becomes something it can draw on.

Step 3: Open the folder and start asking

Open the folder and start the AI assistant from inside it, and its answers are drawn from that folder alone. If you’d rather stay where you are, right-click the folder and add it to the assistant’s context, which has the same effect.

Ask the way you’d ask a colleague who had read everything. “Summarize how the scope changed between the first and final brief.” “Which documents mention the March deadline, and do they agree?” “List every open question raised across these files.” Questions that compare documents or pull a theme from several of them are where this approach earns its time.

One subscription covers models from the major labs, including Gemini, OpenAI, Claude, DeepSeek, Grok and Qwen, so you can pick the model that suits the question without moving your files to another app. The free plan includes limited AI use.

Step 4: Check the sources and keep the conversation going

Answers cite their sources with clickable references, so you can open the passage the assistant relied on and read it in context. That matters most when the answer will go into something else, such as a report or a reply to a client. The document chat page shows how these references appear.

Follow-up questions work the way they would in conversation. If an answer mentions a risk in the third document, ask what the other documents say about it, or ask for the exact wording. Narrowing a question through a few turns is usually faster than writing one perfect prompt.

Step 5: Keep the answers and act on them

A useful answer can be saved as a note, so the summary or comparison you asked for sits in Fabric next to your other work instead of in a chat history. Over a project, those saved answers build into a record of what you found and where you found it.Notes and docs in Fabric can be edited and shared like any other file.

The assistant can also edit documents directly. Ask it to tighten a section, apply a change across a draft, or add a missing point, and each change appears for you to accept or reject. Your original stays as it was until you approve an edit.

Frequently asked questions

Can I ask one question across several documents at once?

Yes. Put the documents in a folder and open the assistant there, or right-click the folder and add it to the assistant’s context. It analyzes everything in that folder to answer.

What file types can I chat with?

PDFs, DOCX, TXT, Markdown, HTML and JSON files, plus notes you write in Fabric. Google Docs and Notion pages come in through connections, which need a paid plan.

Do I have to upload my files every time I start a new chat?

No. Files stay in Fabric, and any new conversation can use them. How to use AI without uploading files every time covers this in more detail.

How do I know the AI’s answer is correct?

Each answer cites its sources with clickable references. Open the reference to read the original passage before relying on the answer.

Which AI models can I use?

One subscription covers models from the major labs, including Gemini, OpenAI, Claude, DeepSeek, Grok and Qwen, among others.

Can the AI rewrite part of my document?

Yes. The assistant can edit documents directly, and you accept or reject each change it proposes.

Can I chat with my documents from ChatGPT or Claude instead?

Yes. The Fabric MCP server lets ChatGPT, Claude and other AI tools search Fabric and read file contents. Sign-in uses OAuth, so there are no API keys to manage.

How does chatting with documents work behind the scenes?

Fabric creates embeddings for each item so search can match by meaning, then the assistant answers from the passages it finds. The general technique is explained in what is retrieval-augmented generation.

Can I chat with my documents on the free plan?

Yes, with limited AI use. Connections to apps like Google Drive need a paid plan. See pricing for the current plans.